| Title | Prepoznavanje redova usjeva na slici primjenom metoda globalne optimizacije |
| Author | Ivan Vidović |
| Mentor(s) | Željko Hocenski
|
| Abstract | U ovoj doktorskoj disertaciji naglasak je stavljen na razvoj metoda za prepoznavanje
redova usjeva na slici. Cilj je razviti metodu koja posjeduje sljedeca svojstva: robusnost
s obzirom na prisutnost korova i sjene, mogucnost prepoznavanja redova usjeva za
razlicite kulture u razlicitim stadijima rasta, mogucnost prepoznavanja razlicitog broja
proizvoljno razmaknutih redova usjeva te mogucnost prepoznavanja ravnih i zakrivljenih
redova usjeva. Razvijene su dvije metode koje su nazvane CRDI metoda i TMGEM
metoda. CRDI metoda zasnovana je na inkrementalnoj metodi za trazenje priblizno optimalne
particije skupa podataka te primjeni DIRECT algoritma za globalnu optimizaciju.
Metoda je pogodna za prepoznavanje ravnih redova usjeva, pri cemu broj redova usjeva
mora biti unaprijed poznat. TMGEM metoda zasnovana je racunanju podudaranja s
predloskom i minimizaciji funkcije globalne energije primjenom dinamickog programiranja.
Funkcija globalne energije, uz podatke sa slike, koristi i prethodno znanje o geometrijskoj
strukturi redova usjeva, cime se postize veca tocnost prepoznavanja redova usjeva.
Predlozena metoda ima mogucnost prepoznavanja proizvoljnog broja redova usjeva za
razlicite kulture u razlicitim stadijima rasta te pri razlicitim geometrijama polja. Nadalje,
predlozena TMGEM metoda robusna je na zakrivljenost redova usjeva. Osim navedenih
metoda, u radu je predlozen novi evaluacijski okvir za usporedbu metoda prepoznavanja
redova usjeva koji ukljucuje bazu slika redova usjeva, postupak rucnog generiranja referentnih
vrijednosti redova usjeva te kriterije za odredivanje tocnosti prepoznavanja redova
usjeva. Na slikama u bazi snimljeni su redovi usjeva kukuruza, celera, krumpira, luka,
suncokreta i soje. Koristenjem predlozenog evaluacijskog okvira moguce je ekasnije i
objektivnije usporediti novo razvijene metode za prepoznavanje redova usjeva s postojecim
metodama. U eksperimentalnoj evaluaciji CRDI metoda je usporedena s IMLD i HT metodom
na umjetno generiranim skupovima podataka. Tocnost prepoznavanja odredena
je primjenom CRHID kriterija, a rezultati pokazuju da CRDI metoda prepoznaje redove
usjeva sa znatno vecom tocnoscu u odnosu na ostale razmatrane metode. TMGEM
metoda testirana je na 225 stvarnih slika redova usjeva iz baze slika te je usporedena s HT, HTT i LR metodom. Tocnost prepoznavanja redova usjeva odredena je koristenjem
CRDA kriterija, a rezultati pokazuju da predlozena metoda znacajno nadmasuje ostale tri
razmatrane metode pri prepoznavanju ravnih redova usjeva te je robusna na zakrivljenost
redova usjeva. |
| Keywords | agricultural automation computer vision crop row detection DIRECT dynamic programming incremental method optimization template matching |
| Parallel title (English) | Image based crop row detection using global optimization methods |
| Committee Members | Robert Cupec (committee chairperson)
|
| Granter | Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek |
| Lower level organizational units | Department of Computer Engineering and Automation
Chair of Computer Engineering |
| Place | Osijek |
| State | Croatia |
| Scientific field, discipline, subdiscipline | TECHNICAL SCIENCES
Computing
Architecture of Computer Systems
|
| UDK | 621.3
APPLIED SCIENCES. MEDICINE. TECHNOLOGY
Electrical engineering |
| Study programme type | university |
| Study level | postgraduate |
| Study programme | Postgraduate doctoral study programme; specializations in: branch Power Engineering and Communications, branch Informatics |
| Study specialization | branch Informatics |
| Academic title abbreviation | dr.sc. |
| Genre | doctoral thesis |
| Language | Croatian |
| Defense date | 2016-05-25 |
| Promoted date | 2017-04-07 |
| Parallel abstract (English) | The topic of this doctoral thesis is the development of image based crop row detection
methods. The main goal of the research is to develop a method which is: highly insensitive
to the presence of weeds and shadows, capable of detecting crop rows of dierent crop
types at dierent stages of growth, capable of detecting straight and curved crop rows
and insensitive to the number and spacing of crop rows. Two methods are proposed
entitled CRDI and TMGEM. CRDI method is based on incremental method of searching
for an approximate globally optimal partition of a set of data points and on the DIRECT
algorithm for global optimization. The method is capable of detecting straight crop
rows, wherein the number of crop rows must be known in advance. TMGEM method
is based on template matching followed by global energy minimization with dynamic
programming technique. For accurate crop row detection, the global energy function
combines image evidence and prior knowledge about the geometric structure of crop rows.
The proposed method is insensitive to the number and spacing of crop rows and is capable
of detecting crop rows of dierent crop types at dierent stages of growth. Furthermore,
the proposed TMGEM method is capable of detecting curved crop rows. A new evaluation
framework is proposed that consists of a crop row image database, manual ground truth
image creation approach and two crop row detection performance measures. The image
database includes images of dierent crop types including maize, celery, potato, onion,
sun
ower and soybean. The proposed evaluation framework enables ecient and objective
comparison of new crop row detection methods with existing ones. The experimental
evaluation of CRDI method includes comparison with IMLD and HT method on synthetic
datasets, based on the proposed CRHID performance measure. The results show that
CRDI method outperforms other considered methods. TMGEM method is evaluated on
a set of 225 real-world crop row images from the image database and it is compared with HT, HTT and LR method. The proposed CRDA measure is used as a performance
measure in the comparison. The results show that TMGEM signicantly outperforms
the other considered methods in straight crop row detection and is capable of detecting
curved crop rows. |
| Parallel keywords (Croatian) | automatizacija u poljoprivredi DIRECT dinamicko programiranje inkrementalna metoda optimizacija podudaranje s predloskom prepoznavanje redova usjeva racunalni vid. |
| Version | accepted version |
| Resource type | text |
| Access condition | Open access |
| Terms of use |  |
| Note | accepted version |
| URN:NBN | https://urn.nsk.hr/urn:nbn:hr:200:880047 |
